AI Security & Safety Engineer
I work on the security and safety of AI systems. That means adversarial testing of agents and guardrails, and the engineering that keeps a compromised model from reaching anything that matters.
My independent work is adversarial and public. I reproduce attacks, publish what held and what did not, and teach the same material to students who will run into it for real.
02 / research
Adversarial work on agentic systems, guardrails and the platforms they run on. The numbers are from the evaluations, and the failures are written up alongside them.
A reproduction of the Morris II zero-click GenAI worm across a mesh of eight tool-using agents, with an LLM judge as the only guardrail. The interesting results are the guardrail's failures, not its catches.
An elicitation harness for agentic misbehaviour. Three constructed scenarios in which an autonomous agent under pressure reaches for coercion, social engineering, or exfiltration. An existence proof, deliberately, rather than a base rate.
An adversarial test suite of 44 prompts across seven attack categories, with a Kenyan PII corpus and a Swahili/Sheng attack corpus, run against commercial AI guardrail products. Products marketed as multilingual performed poorly outside English.
Six hands-on Colab labs where students build a small AI app, break it with real techniques, then secure it and re-run the same attacks to watch the defences hold. Taught across Kenyan universities.
An enterprise CTI platform processing over 100,000 daily events from OSINT and dark web sources, with an analysis engine on Vertex AI for automated summarisation, MITRE ATT&CK mapping and TTP classification.
A security workshop built as a playable Snake game. Co-built for Decode 4.0 and still the most-forked thing I have written.
03 / experience
Five years building and defending platforms at telecom scale, then moving the same work onto AI systems.
Safaricom PLC
Leading the AI security function across two pillars: securing the AI systems the business builds and procures, and applying AI to cyber defence.
Safaricom PLC
Architected and built the threat intelligence platform, processing over 100,000 daily events from OSINT and dark web sources with an analysis engine on Vertex AI.
Safaricom PLC
Led development of the Rich Communication Services platform and the abuse and fraud controls on it, and built fault-tolerant telecom APIs for FTTH, 4G and 5G networks.
Sawasawa Ltd
Built Python, Django and JavaScript web applications and REST APIs on a microservices architecture.
Centum Learning Ltd
Built e-learning and mobile training platforms for corporate onboarding.
04 / teaching
Reading about an attack and having broken something yourself are not the same thing. The labs are built around the second one.
Running hands-on labs for university students across nine regions through 2026 and 2027, with a team of four running them in parallel.
Presented CONTAGION live: agentic worm propagation across a mesh of eight tool-using agents, and where the LLM judge guarding it fails.
Taught the Break & Secure red team labs: build an AI app, attack it, secure it, then re-run the same attacks against the fix.
Co-built the security workshop as a playable game. 42 forks, meaning people took it away and ran it themselves.
05 / writing
Longer notes on the systems I build and break.
I built a 44-prompt adversarial suite with a Kenyan PII corpus and a Swahili/Sheng attack corpus, and ran it against commercial guardrail products sold as multilingual. The best one blocked 21 of 44, and the configuration that scored higher was the worse product.
Learn how to integrate vector search and AI into your Next.js applications for intelligent, context-aware features. A practical guide with real-world examples.
How we architected a CTI platform to process 100K+ daily events using Python, PostgreSQL, vector search, and AI/ML. A deep dive into distributed systems, performance optimization, and real-time threat analysis.
06 / contact
Happy to talk about AI security architecture, guardrail evaluation, or bringing the builder labs to a university or team.
Nairobi, Kenya